fal video models

# Fix list: fal video models

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/fal-video, the October 2026 research run, assessed 8 October 2026. Grade B, 68.9 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on fal video models: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Payments & pricing, 20 out of 100, up to 10 more on the total

Why it scored 20: No x402, MPP or L402 in the docs, pricing page or terms (0). Per-model prices are public without a login on fal.ai/pricing and on every model page and llms.txt, and a pricing API returns unit prices by endpoint ID (20). No standing free tier. The terms and FAQ mention promotional credits and grants with their own expiry, and the terms require credit bought in advance (0). A person signs up in a browser and buys credit, and the MCP relay needs browser sign-in (0).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 2. Security & auth, 63 out of 100, up to 6.5 more on the total

Why it scored 63: Keys carry one of two scopes, API (call models) or ADMIN (CLI, app management and admin platform APIs), are shown once at creation and are managed in the dashboard. Rotation and expiry aren't documented. The MCP relay uses OAuth with browser sign-in, and its scopes aren't documented (25). The API scope is the least-privilege mode. There is no read-only key, and the MCP docs state that the server has no server-side approval gate before a paid generation (10). Returns generated media and fal's own catalogue text (10). Request payloads are kept 30 days in the dashboard history, and usage and analytics APIs report per-request spend (10). `https://fal.ai/.well-known/security.txt` returned 404 on 8 October 2026 and no disclosure policy or bug bounty was found. The enterprise page claims SOC 2 certification, the DPA of 31 July 2026 has a data security exhibit, and the trust centre at trust.fal.ai renders only in a browser, so the report and its scope were unread (8).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 3. Reliability, 77 out of 100, up to 4.6 more on the total

Why it scored 77: Graded as a hosted service on the queue API. Instatus page at status.fal.ai with components for Model API, Serverless API, Official Models and the dashboards, and monthly uptime figures (20). Two notices from August to 8 October 2026, a 30-minute major outage of synchronous calls to `fal.run` on 4 September and 18 minutes of degraded performance on 29 September. Neither reached an hour, and the page showed nothing before August, so 10 to 31 July is unread (20). Concurrency is published, 2 for a new account rising to 40 self-serve with paid invoices over four weeks, and queued requests are never rejected (15). A 429 carries type `concurrent_requests_limit` and `X-Fal-needs-retry: 1`, the docs give backoff of 1, 2, 4 and 8 seconds, the queue retries 503, 504 and connection errors up to 10 times, and backup domains are documented. There is no idempotency key, and one page says concurrency requeues have no maximum while another says 10 (12). The enterprise page mentions SLA guarantees but no SLA document is published (0). The queue API is generally available (10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 4. Transparency & trust, 60 out of 100, up to 3.5 more on the total

Made of editorial 53, provenance 66.

Why it scored 60: Closed platform with published terms (8 September 2026) and API Services supplemental terms. The clients are MIT and the catalogue marks each model's `license_type`. The terms warrant nothing about rights in output, and a clause bars scripts that integrate with the services without written authorisation, which sits oddly with an API product (13). Data handling is documented and mostly agrees. Request payloads are kept 30 days with an opt-out header and a delete API, the API Services terms say client content isn't used to train or develop fal's products except on models marked Pending Enterprise Ready, and the privacy policy of 22 July 2026 deletes account data 30 days after closure. The terms still let fal use usage data to develop AI models, and content sent to a third-party model goes to that third party (22). 22 text-to-video endpoints are marked deprecated in the catalogue, with no removal dates, no written policy and no changelog entries (8). The DPA names a subprocessor list at trust.fal.ai/subprocessors with 15 days' notice of changes, but that page renders only in a browser, and the privacy policy says processing is in the United States and other countries (10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Domain age: fal.ai, registered 2020-11-13 (5 years) (11 of 15)
- Endpoint on the vendor's domain: queue.fal.run is not on fal.ai (0 of 15)
- Terms of service: read, states 6 of the 7 things a reader expects, and has 2 clauses that cost points (5.1 of 10)
- security.txt: not found (0 of 10)

## 5. Agent ergonomics, 81 out of 100, up to 3.1 more on the total

Why it scored 81: Graded on the queue API, with the hosted MCP server noted. Results are small JSON with a file URL, and a model's llms.txt is 6 to 8 KB loaded on demand. The MCP server has 11 tools with schema and pricing lookups (20). The catalogue API takes `category`, `limit` and a cursor, queue status reports position, webhooks replace polling, and most models take duration and resolution, though names and types differ (18). Errors carry a machine-readable `error_type`, the same value in `X-Fal-Error-Type`, and guidance on which to retry (18). No idempotency key. Server-side retries, `X-Fal-No-Retry`, a cancel endpoint and no charge for 5xx failures reduce the cost of a failed call, but a repeated submit is a second billable job, and the MCP docs don't mention readOnlyHint or destructiveHint (10). Few required parameters, and official Python and JavaScript clients plus Swift, Kotlin and Dart (15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 6. Schema & documentation, 89 out of 100, up to 1.8 more on the total

Why it scored 89: Per-endpoint OpenAPI 3.0.4 documents for the queue API, for example `fal-ai/kling-video/v3/pro/text-to-video`, and an OpenAPI 3.1 document for the platform APIs (25). A docs llms.txt, Markdown twins of docs pages and an llms.txt per model with schema, defaults and price (10). Model pages state what each model does but rarely when not to use it, and the Seedance 2.5 catalogue text says up to 720p while its schema lists 1080p (12). Inputs are typed with enums, ranges and defaults, but `duration` is a string enum on Kling, a string with a unit on Veo 3.1 and an integer on Wan 3.0, and Kling's rule that one of `prompt` or `multi_prompt` is needed sits in prose with a required example of `{}` (13). Examples in cURL, Python and JavaScript on every model, a model error reference with typed errors such as `content_policy_violation` and `video_duration_too_long`, and a request error reference with 13 types (14). Versioned endpoint IDs and a dated public changelog (15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 7. Maintenance & community, 81 out of 100, up to 1.7 more on the total

Why it scored 81: `fal-client` 1.0.3 reached PyPI on 21 September 2026, 17 days before the check, and the catalogue shows video endpoints added on 1 and 5 October (30). Six dated changelog entries from 3 August to 14 September and three Python client releases since 19 August (20). Public changelog, Discord and a support address, with response times not checked (10). Python client current at 1.0.3, JavaScript client at 1.10.1 on npm with the repository pushed on 7 October 2026, and Swift, Kotlin and Dart clients documented (13). `fal-js` has 19 open issues and pull requests, and its CI wasn't read (8).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- unchecked: trust.fal.ai and trust.fal.ai/subprocessors render only in a browser, so the SOC 2 report, its scope and the subprocessor list were unread.
- unchecked: the status history showed August to October 2026 only, so 10 to 31 July of the 90-day window is unread.
- unchecked: the MCP relay at `https://mcp.fal.ai/mcp-relay` needs OAuth sign-in, so tool definitions, annotations and scopes are taken from the docs, not the server.
- unchecked: the official MCP registry. A search for fal returned 30 unrelated servers on its first page and no fal entry.
- unchecked: the pricing API at `/v1/models/pricing` needs a key, so prices come from model pages. Nine models were read, not every endpoint.
- The pricing page lists Kling v3 Pro at $0.14 a second while the model page gives $0.112 without audio and $0.168 with audio. Which one is billed wasn't established.
- The catalogue marks 22 text-to-video endpoints deprecated without removal dates, so whether they still answer wasn't established.
- MiniMax H3 Max prices on its model page are promotional until 15 October 2026. The unit price recorded is the stated price after that date.
- The docs llms.txt opens with a block headed Agent Instructions addressed to AI models. It was read as data and not acted on.
- The existing fal-image and fal-music listings record no DPA, no training statement and no error reference. All three exist today and those listings want a recheck.
- Response times of fal's support channels and the release date of `@fal-ai/client` 1.10.1 weren't checked.

## Weaknesses

- Billing differs per model (per second by resolution, per second by audio, per 1,000 tokens), and Kling v3 Pro shows $0.14 on the pricing page against $0.112 or $0.168 on its model page
- No standing free tier, no machine payment protocol, and sign-up and credit purchase need a person in a browser
- 22 text-to-video endpoints, Sora 2 and Veo 3 among them, are marked deprecated in the catalogue with no removal date or changelog entry
- No idempotency key, so a repeated submit starts a second billable job
- No security.txt, and the trust centre with the SOC 2 material and subprocessor list renders only in a browser
- The terms of service bar scripts that integrate with the services without written authorisation, and bar robots and data gathering

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Submit to `queue.fal.run`, keep the `request_id`, then poll `status_url` or set `fal_webhook`. Never resubmit to check progress, because each submit is a new billable job
- Read `https://fal.ai/models/<endpoint-id>/llms.txt` before a call. `duration` is the string `"5"` on Kling, `"8s"` on Veo 3.1 and an integer on Wan 3.0
- Set `generate_audio` on purpose. It defaults to true and raises the Kling v3 Pro price from $0.112 to $0.168 a second and Veo 3.1 from $0.20 to $0.40
- Check `metadata.status` in `https://api.fal.ai/v1/models` before pinning an endpoint. Deprecated endpoints carry no removal date
- Use an API-scope key, never an ADMIN key. A new account runs 2 requests at once, so long video jobs queue behind each other
- The MCP relay has no server-side approval gate. Call `get_pricing` and get the owner's approval before `submit_job`

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.